Advanced Machine Learning

Lecture 1: Introduction
Sandjai Bhulai
Vrije Universiteit Amsterdam

[email protected]
6 September 2023
 Motivation

2 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Motivation

3 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Motivation

4 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Motivation

5 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Motivation

6 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Motivation

7 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning

8 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning
What is machine learning?
▪ Arthur Samuel (1959): eld of study that gives computers the
ability to learn without being explicitly programmed

▪ Tom Mitchell (1998): a computer program is said to learn

from experience E with respect to some task T and some
performance measure P, if its performance on T, as
measured by P, improves with experience E

9 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fi
 Machine learning

10 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning examples
▪ Database mining
> Web click data, medical records, biology

▪ Applications that cannot be programmed by hand

> Autonomous vehicles, OCR, NLP, computer vision

▪ Self-customizing programs
> Amazon, Net ix product recommendations

▪ Understanding human learning / behavior

> Brain, real AI

11 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fl
 Machine learning examples

12 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning examples
▪ Database mining
> Web click data, medical records, biology

▪ Applications that cannot be programmed by hand

> Autonomous vehicles, OCR, NLP, computer vision

▪ Self-customizing programs
> Amazon, Net ix product recommendations

▪ Understanding human learning / behavior

> Brain, real AI

13 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fl
 Machine learning examples

14 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning examples
▪ Database mining
> Web click data, medical records, biology

▪ Applications that cannot be programmed by hand

> Autonomous vehicles, OCR, NLP, computer vision

▪ Self-customizing programs
> Amazon, Net ix product recommendations

▪ Understanding human learning / behavior

> Brain, real AI

15 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fl
 Machine learning examples

16 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Machine learning examples
▪ Database mining
> Web click data, medical records, biology

▪ Applications that cannot be programmed by hand

> Autonomous vehicles, OCR, NLP, computer vision

▪ Self-customizing programs
> Amazon, Net ix product recommendations

▪ Understanding human learning / behavior

> Brain, real AI

17 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fl
 Machine learning examples

18 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview

Advanced Machine Learning

 Course overview: objectives
▪ Understand the capabilities and the limitations of machine
learning
▪ Implement machine learning algorithms in Python
▪ Know relevant machine learning algorithms for both
supervised and unsupervised learning problems
▪ Select the right machine learning models for real-world use
cases
▪ Understand when to apply online learning, reinforcement
learning, and deep learning
▪ Interpret the outcomes of machine learning algorithms

21 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: grading
▪ Course value: 6 ECTS (a full month of work)
▪ Course consists of:
> Written exam (weight: 0.9)
> Computer assignments (weight: 0.1)
▪ Both parts have to be passed with at least a 5.5
▪ Written assignments for exam preparation
▪ Q&A sessions on Wednesday for written assignments
▪ Teaching assistant: Mathijs Pellemans
▪ Slack workspace

22 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: books
▪ Book presents mathematical
theory behind machine learning
▪ Very mathematical
▪ Lectures will try to explain the
mathematics

23 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: books
▪ Book presents concepts behind
deep learning
▪ Not mathematical
▪ Lectures will try to explain the
mathematics

24 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: books

25 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: program
▪ Regression models (linear, logistic, etc.)
▪ Neural networks (NNs)
▪ Convolutional NNs
▪ Recurrent NNs
▪ Hidden Markov models
▪ Reinforcement learning

26 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Course overview: software

27 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Preliminaries

Advanced Machine Learning

 Linear algebra
▪ A matrix A has elements Aij where i indexes rows and j indexes
columns

Properties of matrices:
▪ AA −1 = A −1A = I
▪ (AB)⊤ = B ⊤ A ⊤
▪ (AB)−1 = B −1A −1
▪ (A ⊤)−1 = (A −1)⊤
▪ (A ⊤)⊤ = A
▪ (A + B)⊤ = A ⊤ + B ⊤
29 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023
 Linear algebra
▪ Symmetric matrix: A⊤ = A
▪ Positive de nite: x ⊤ Ax > 0 for all non-zero x and A
symmetric

▪ All vectors in this course will be column vectors

▪ Recall, if A ∈ ℝm×n is a matrix and x ∈ ℝn a vector, then the
n

∑
i-th element of Ax is given by (Ax)i = aij xj
j=1

30 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fi
 Vector calculus
▪ Let a ∈ ℝn and b ∈ ℝn be a vector, and x ∈ ℝ a scalar.
Then

( ∂x )
∂a ∂ai
=
i
∂x

Derivatives with respect to a vector are de ned by

( ∂a )
∂x ∂x
▪ scalar-vector =
i
∂ai

( ∂b )
▪ vector-vector ∂a ∂ai
=
ij
∂bj

31 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fi
 Vector calculus
n ∂ ⊤ n
▪ Example: let a ∈ ℝ and x ∈ ℝ . What is (x a)?
∂x

▪ Solution: this is a scalar-vector derivative, thus

n
∂ ∑j=1 aj xj
( ∂x )
⊤ ⊤
∂x a ∂x a
= = = ai
∂xi ∂xi
i
∂ ⊤ ⊤ ∂ ⊤
▪ Thus (x a) = a , similarly (a x) = a⊤
∂x ∂x

32 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Vector calculus
▪ Another example. Let b ∈ ℝm, A ∈ ℝm×n, and x ∈ ℝn
∂b⊤ Ax
▪ What is ?
∂x

▪ Solution: this is a scalar-vector derivative, thus

n m
∂ ∑k=1 ∑j=1 bjajk xk m m

( ∂x )
⊤ ⊤
∂b Ax ∂b Ax
aij⊤bj = (A ⊤b)i
∑ ∑
= = = bjaji =
∂xi ∂xi j=1 j=1
i

∂b⊤ Ax
▪ Hence, = b⊤ A
∂x
33 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023
 Vector calculus
∂x
▪ =I
∂x

∂Ax
▪ =A
∂x

∂x⊤ A
▪ = A⊤
∂x

∂x⊤ Ax
▪ = x⊤(A + A ⊤) and 2x⊤ A if A is symmetric
∂x

∂2x⊤ Ax ⊤
▪ = A + A and 2A if A is symmetric
∂x∂x ⊤

34 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Vector calculus
∂x
▪ =I
∂x

▪ Proof:

( ∂x ) {1,
∂x ∂xi 0, if i ≠ j,
= =
∂xj if i = j
ij

35 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Vector calculus
∂Ax
▪ =A
∂x

▪ Proof:

( ∂x )
∂Ax ∂(Ax)i ∂ ∑k=1 aik xk
= = = aij
ij
∂xj ∂xj

36 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Vector calculus
∂x⊤ Ax
▪ = x⊤(A + A ⊤)
∂x

▪ Proof:

∂ ∑k=1 ( ∑l=1 akl xl) xk

n n
∂x⊤ Ax
∑ ∑
= = aki xk + ail xl + 2aii xi =
∂xi ∂xi k≠i l≠i
n n
ail xl = (A ⊤x)i + (Ax)i
∑ ∑
aki xk +
k=1 l=1

37 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Vector calculus
⊤ ⊤
n ∂a xx b
▪ If a, b, x ∈ ℝ , then = x⊤(ab⊤ + ba⊤)
∂x

▪ Proof:

n n
⊤
∂a xx b ⊤ ∂ ∑k=1 ak xk ∑l=1 xlbl n n

∑ ∑
= = bi ak xk + ai xlbl =
∂xi ∂xi k=1 l=1

bia⊤x + aix⊤b = bia⊤x + aib⊤x = (bia⊤ + aib⊤) x

38 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Normal distribution
▪ Univariate Normal distribution

{ 2σ 2 }
2 1 1 2
(x | μ, σ ) = exp − (x − μ)
(2πσ 2)1/2

β = 1/σ 2

(x | μ, σ 2) > 0
∞

∫−∞
(x | μ, σ 2) dx = 1
𝒩
𝒩
39 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023
𝒩
 Normal distribution
▪ Univariate Normal distribution

{ 2σ 2 }
2 1 1 2
(x | μ, σ ) = exp − (x − μ)
(2πσ 2)1/2

∫−∞
[x] = x (x | μ, σ 2) dx = μ
∞

∫−∞
[x 2] = x 2 (x | μ, σ 2) dx = μ 2 + σ 2

var[x] = [x 2] − [x]2 = σ 2
𝒩
𝔼
𝔼
𝒩
𝒩
40 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023
𝔼
𝔼
 Normal distribution

41 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

 Normal distribution
▪ Multivariate Normal distribution

{ 2 }
1 1 1 ⊤ −1
(x | μ, Σ) = exp − (x − μ) Σ (x − μ)
(2π) D/2
|Σ| 1/2

▪ μ is a D-dimensional mean vector

▪ Σ is a D × D covariance matrix (symmetric, positive de nite)
▪ | Σ | is the determinant of Σ
𝒩
42 Sandjai Bhulai / Advanced Machine Learning / 6 September 2023

fi
 Normal distribution
([0] [0 1]) ([1] [0.8 1 ])
0 1 0 0 1 0.8
, ,

43
𝒩
𝒩
Sandjai Bhulai / Advanced Machine Learning / 6 September 2023